Model-Based Design of Runtime Adaptation Strategies

Designing and implementing embedded computing systems is more challenging than non-embedded systems because of the need to interact closely with a changeable physical universe that operates in real time, and because of the extra constraints imposed by the packaging and environmental concerns for many embedded systems. In addition, the challenges posed by distributed systems architectures and composite implementations of stand-alone components (which are already prevalent in non-embedded environments) have begun to become commonplace consequences of the requirements for new embedded solutions. There are especially challenging issues surrounding the design and implementation of distributed embedded systems with managed quality of service (QoS) properties when those systems must adapt to the changes in both the computational and physical universes within which these systems must operate. To date, research has concentrated on constructing runtime system abstractions and mechanisms that can adaptively meet the design requirements. However, using these abstractions and mechanisms requires highly skilled individuals with strong intuitions about both the needs of the domain and the ways to manipulate the various dimensions contributing to the managed QoS behavior. There are few design paradigms to guide the design or construction of this sort of dynamic, adaptive, managed runtime behavior. There is a compelling need for applying design-time methodologies to develop and control these runtime adaptations systematically, and along the way to establish appropriate interchanges and interfaces between the designtime tools and their runtime counterparts. Under the auspices of the DARPA MoBIES program, we are investigating the application of model integrated computing (MIC) to the problems of designing, customizing, parameterizing, and managing the adaptive runtime characteristics of distributed realtime embedded (DRE) applications. In particular, we are seeking to utilize and augment the Generic Modeling Environment (GME) MIC tool to design runtime control capabilities as provided by the Quality Objects (QuO) adaptive QoS middleware framework. GME uses domain models and advanced user/designer interfaces to manipulate elements in the domain space that are intelligently interpreted by the models to produce effective designs and code elements representing that design [4]. QuO provides a set of extensions to existing off-the-shelf middleware, including CORBA and Java RMI, specification languages, and a runtime system to support QoS awareness and management of, and adaptation to, dynamic conditions [18].